MCP Server - Vector Search

@omarguzmanm

リポジトリ

基本情報

カテゴリ

メモリとナレッジ

トランスポート

stdio

作者

omarguzmanm

MCP Server - Vector Search について

MCP Server to improve LLM context through vector search.

#neo4j #mcp #knowledge-graph #llm

MCP Server - Vector Search の設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "mcp-server-vector-search": {
      "command": "uv",
      "args": [
        "venv"
      ]
    }
  }
}

MCP Server - Vector Search のツール

ツール一覧はまだありません

この掲載には表示できるツール一覧がまだありません。サーバーにツールがないという意味ではありません。対応機能はプロジェクトのドキュメントをご確認ください。

ツールは README から自動的に抽出されます。メンテナーは ## Tools という見出しの下に記載することで、このタブに反映できます。

概要

What is 🔍 MCP Server - Vector Search?

A Model Context Protocol server built with FastMCP that combines Neo4j’s graph database with vector search using embeddings. It enables intelligent semantic search across a knowledge graph through natural language queries, designed for MCP clients such as Claude AI.

How to use 🔍 MCP Server - Vector Search?

After cloning the repository, create a virtual environment with uv, install dependencies (fastmcp, neo4j, openai, python-dotenv, sentence-transformers, pydantic), configure a .env file with Neo4j credentials and an optional OpenAI API key, and create a vector index in Neo4j. Launch the server with python main.py. The server exposes one tool: vector_search_neo4j(prompt), which converts a natural language query into an embedding and searches the vector index for semantically similar nodes.

Key features of 🔍 MCP Server - Vector Search

Converts natural language queries into 1536‑dimensional embeddings via OpenAI.
Searches a Neo4j vector index for semantically similar nodes.
Returns ranked results with similarity scores.
Built on FastMCP for minimal overhead and MCP protocol compliance.
Uses uv for 10–100x faster dependency resolution.
Supports fallback to a local all-MiniLM-L6-v2 embedding model.

Use cases of 🔍 MCP Server - Vector Search

Semantic document retrieval from a Neo4j knowledge graph.
Finding contextually relevant graph-connected information using plain language.
Integrating intelligent search into MCP‑compatible AI assistants (e.g., Claude Desktop).
Building RAG‑style applications that combine graph traversal with vector similarity.

よくある質問

What are the runtime requirements?

Python 3.8+, Neo4j 5.0+ with the APOC plugin, and either an OpenAI API key (for the default 1536‑dimension embeddings) or the `sentence-transformers` library for a local fallback model.

Where does the data live?

All data – nodes, their `embedding` properties, and the vector index – is stored inside a Neo4j database. The server only reads from and writes to that database.

Is an OpenAI API key required?

No. If `OPENAI_API_KEY` is not set in `.env`, the server falls back to the local `all-MiniLM-L6-v2` model from `sentence-transformers`.

What vector index must exist in Neo4j?

A vector index named `embeddableIndex` on nodes labeled `Document` with property `embedding`, dimension 1536, and cosine similarity. If using the local fallback model, adjust the dimension accordingly.

How do I troubleshoot a missing vector index?

Use Cypher `SHOW INDEXES` to verify existence, and re‑create it with the `CREATE VECTOR INDEX` command shown in the Quick Start.

「メモリとナレッジ」の他のコンテンツ

agentleFS

One permissioned folder your team and your AI agents both log into.

UmmahAPI

Yousuf Kazmi

Islamic source data for assistants: Quran, hadith, tafsir, duas, prayer times and qibla. Answers come from real sources with citations instead of recalled text, so verses and hadith are not hallucinated. 36,000+ hadiths